The tech giant unveiled its third-generation custom silicon, aiming to reduce reliance on Nvidia and cut energy costs for AI inference tasks.
Key points
- Microsoft unveiled the Maia 300, its third-generation custom AI chip, on Monday.
- The company plans to make Maia chips the primary inference system for its products by 2027.
- Microsoft expects the chip to deliver 40% more tokens per megawatt than leading GPU systems.
Microsoft introduced the Maia 300, its third-generation custom artificial intelligence chip, on Monday. The company announced it will make this silicon the primary system for running AI in its products starting in 2027.
What happened
Microsoft unveiled the Maia 300 on Monday, marking the third generation of its custom AI hardware. According to The Next Web, the company plans to deploy these chips in gigawatt-scale data centers starting next year, which could involve more than a million units by 2027. Microsoft stated that the Maia family will become the primary inference system for its products and services by 2027. Scott Guthrie, Microsoft’s executive vice president for cloud and AI, said in a blog post that the chip delivers 40% more tokens per megawatt compared to leading GPU-based systems. The Next Web reported that each rack holds 96 chips and draws about 167 kilowatts, delivering up to seven times the output of its predecessor, the Maia 200.
Reuters noted that Nvidia’s latest GB300 racks hold 72 GPUs and draw about 142 kilowatts by comparison. The chip uses TSMC’s N2P process and eight stacks of HBM4 memory, split into smaller chiplets for computing and data transfer. The Verge noted that this move reduces Microsoft’s dependence on Nvidia and AMD. The Maia 200, launched in January, currently supports Microsoft Copilot and will support GitHub Copilot and Microsoft Foundry by the end of 2026, according to Computerworld. Microsoft plans to share more details at its Ignite conference in San Francisco next month.
Why it matters
Custom silicon allows cloud providers to tailor hardware specifically for their workloads, often improving efficiency over general-purpose chips. Inference refers to the process of running trained AI models to generate responses, which consumes significant power. By optimizing for tokens per watt, companies can lower operational costs and reduce energy consumption. Shifting to in-house chips also reduces supply chain risks associated with relying on a single external vendor for critical infrastructure components.
The fix
- Monitor vendor announcements for deployment timelines of new infrastructure components.
- Review current hardware contracts to assess potential cost savings from efficiency upgrades.
- Evaluate supply chain dependencies on single-source hardware providers.
- Plan capacity expansions based on projected energy consumption metrics.
Why it matters
Organizations should understand how custom silicon impacts service costs and performance. Teams usually analyze energy efficiency metrics when planning large-scale deployments. Check whether your current providers are shifting to in-house hardware to anticipate changes in service pricing or availability. Evaluate how vendor lock-in affects your long-term infrastructure strategy. Consider the environmental impact of high-power computing when selecting cloud partners. That suggests a trend toward greater vertical integration in the cloud industry.
Background: Supply chain attacks
Verify code signatures and hash values before installing any update. Monitor for unexpected outbound traffic or permission requests. If anomalies appear, isolate the system immediately and revert to a known good state. Treat third-party updates as untrusted until proven otherwise.
Read the full guide: Supply Chain Attack Warning Signs and Response Steps
What to do and how to stay safe: Microsoft
- Review your organization's cloud contracts for clauses related to hardware changes.
- Assess the energy efficiency claims of your current AI service providers.
- Monitor vendor roadmaps for updates on custom silicon deployments.
- Evaluate the impact of hardware shifts on your application performance.
Step-by-step guide: What is password policies and how they protect systems
General security guidance from the Crypt Weekly newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
When will Microsoft start using the Maia 300 chip?
Microsoft plans to deploy the Maia 300 after the turn of the year, with large-scale deployments starting in 2027.
How does the Maia 300 compare to GPU systems?
Microsoft expects the Maia 300 to deliver 40% more tokens per megawatt compared to leading GPU-based systems.
What AI services currently use Microsoft's custom chips?
The Maia 200 currently supports Microsoft Copilot and will support GitHub Copilot and Microsoft Foundry by the end of 2026.
